PURPOSE: To enable decision close to the thinking of a specialist based on logical knowledge by adding a means to change the input/output characteristic of a specified unit and/or the weight of a specified link according to the output of the unit.

CONSTITUTION: At first, an output from the all or one part of a unit U is fetched according to an arrow A. At such a time, the output is a result accord ing to the input/output characteristic applied as an event in advance by learning. As a logical unit, how to pay attention for a human being is described in advance and the unit, to which any logical meaning is applied, shows any output corresponding to the output result of a neural circuit network arithmetic unit itself. Then, the output is fed back to each unit and link part according to an arrow F. When this fed-back output is observed, the input/output characteris tic of the unit or the strength (weight coefficient) of the link is changed accord ing to the amount of this output and a pattern recognition characteristic is changed. Thus, the decided result is fed back from the logical knowledge to the pattern recognition and made much closer to the specialist.